Automotive · R&D project · Peer-reviewed paper
Automotive display panels: multiview AOI with a YOLOv8n-based detector reaches 99.2% panel-level recall in production-line validation
| Company / plant | Shantou University with Shantou Goworld Display Technology (Guangdong Provincial Key Laboratory of Automotive Display and Touch Technologies), China |
|---|---|
| Defect type | Surface defects on display panels, often obscured by reflections, dust, moiré and viewing angle |
| Part | Automotive display panels |
| Material | Not stated in source |
| Vendor | Research system (YOLO-OG, improved YOLOv8n) |
| Camera / lighting | Multiview image acquisition with dark-field dust suppression and region-of-interest preprocessing |
| Training data | Self-built automotive display panel defect dataset (not public, industrial confidentiality); transfer tests on the public PKU MSD dataset |
| Reported result (quoted) | In production-line validation the system "achieved 99.2% panel-level recall, exceeding that of manual inspection, and produced half the false positives of conventional AOI under the same imaging workflow, with a mean station time of 1.57 s across the defective panels." On the dataset, YOLO-OG improved precision by 7.6, mAP@0.5 by 5.7 and mAP@0.5:0.95 by 2.1 percentage points over the YOLOv8n baseline. |
| Notes | Sensing and Imaging (Springer Nature), published Nov 13, 2026. Research project; authors from Shantou University's College of Engineering and one co-author from Shantou Goworld Display Technology. The paper does not say which production line was used for validation. Authors thank engineers from the Machine Vision Department of the Shantou Institute of Light Industry Equipment. |
| Defect categories | Surface scratches/dents |
| Source | https://link.springer.com/article/10.1007/s11220-026-00926-0 |
Results are quoted from the source. Vendor-published results are vendor claims and were not independently verified. Spot an error? Contact us.